Bridging Police and Communities with Data


Yesterday, President Barack Obama returned to his home city of Chicago to address the International Association of Chiefs of Police (ICAP), focusing his remarks on sentencing reform, gun safety, and reducing tension between police and communities in light of recent, high-profile events in Ferguson, Charleston, and other cities. Concurrent with his appearance, the White House released an update on their Police Data Initiative -- an effort launched this spring with the goal of using data to reduce crime, increase transparency and accountability, and build community trust. One of the projects under the PDI umbrella started this summer at the Data Science for Social Good Summer Fellowship, where one team partnered with the Charlotte-Mecklenburg Police Department to improve their "early intervention system" (EIS) for avoiding adverse police incidents.

Currently, most of the EIS software used by police departments uses thresholds to flag officers at risk of a future incident, with events such as citizen complaints, accidents, and use of force added up until they pass a chosen number where their supervisor is notified and intervention is applied. DSSG fellows Samuel Carton, Kenneth Joseph, Ayesha Mahmud, and Youngsoo Park, with technical mentor Joe Walsh and project manager Lauren Haynes, drew upon more advanced statistical methods to create a new system, using data from internal affairs, arrest reports, dispatches, and other sources provided by CMPD to make more accurate predictions. 

In a post for the DSSG website, the team describes their approach and their preliminary results: a system that flags more high-risk officers while reducing false positives that can waste a department's resources. 

Using relatively straightforward tools from the machine-learning toolkit, we achieved significant improvements both in flagging officers who have adverse interactions and not flagging the rest over the existing system...our models can flag more high-risk officers (75 more than the current system) while flagging fewer low-risk officers (180 fewer than the current system).

Though the summer and the fellowship has ended, the project continues at the Center for Data Science and Public Policy, a researcher center of the CI and the Harris School of Public Policy. DSAPP researchers are working with additional police departments to improve their model and test it in new environments, with the hope of eventually opening up the system to departments nationwide. 

Written By: